Segmentation of Brain Tumors in CT Images: A Comparative Analysis Using Convolutional Neural Network (CNN) Techniques and Support Vector Machine (SVM)

Brain tumor segmentation using CT scans is essential for diagnosis, therapy planning, and progression tracking. This study compares SVM and CNN methods for segmenting brain tumors in CT images. These methods will be assessed for tumor boundary accuracy. To assure uniformity and quality, this study employed preprocessed CT scans from brain tumor patients. CNN uses deep learning to automatically learn hierarchical features during training, while SVM uses handcrafted image features. SVM uses texture analysis, intensity thresholding, and shape descriptors to train the classifier to discriminate tumor and non-tumor regions. The CNN model learns features directly from raw pixel data, using deep convolutional layers' hierarchical nature to capture complex patterns and spatial correlations in images. Both models are split-sampled, trained, validated, and assessed for robustness and generalizability. Performance indicators including accuracy, sensitivity, specificity, precision, and Dice similarity coefficient quantify segmentation results. To examine segmentation results' clinical relevance, professional radiologists do qualitative assessments. Deep learning techniques can be used in medical image analysis, since the CNN model surpasses the SVM model in segmentation accuracy and resilience. SVM has lesser performance but better computing economy and interpretability, which may be useful in some therapeutic scenarios. This work sheds light on the pros and cons of SVM and CNN methods for brain tumor segmentation in CT images, helping physicians create more accurate and efficient tools. CNNs are excellent tools for this endeavor, but classical machine learning methods have a place, especially in cases where computational resources are restricted or model interpretability is important.